【学术报告】追求绿色、颗粒化的机器学习:联邦学习、知识迁移和知识蒸馏方面的进展
发布人:赵振华  发布时间:2022-09-06   浏览次数:10

中文版:

题目:追求绿色、颗粒化的机器学习:联邦学习、知识迁移和知识蒸馏方面的进展

人:Witold Pedrycz 院士


工作单位:阿尔伯塔大学(加拿大)

报告题目:追求绿色、颗粒化的机器学习:联邦学习、知识迁移和知识蒸馏q前沿进展

报告时间:2022910日(周六)10:00-11:00

报告链接:https://meeting.tencent.com/dm/ulnDpkMZP8CA (腾讯会议)(会议号:482 151 152

内容摘要:

  绿色机器学习(亦称绿色人工智能)最近已成为智能系统领域中一项有趣且面向应用的研究方向,它通过分析计算开销(和相关的碳足迹)、可解释性和鲁棒性等,对机器学习架构和学习方案的设计实践进行整体多标准评估。在现实环境中,数据只能在本地可用,而在个别数据岛之外的使用有严格的限制。在解决全球数据分析和全球模型开发问题时,这些限制构成了真正的概念和算法上的挑战,在该领域的许多追求都属于联邦学习的范畴。首先,本报告将讨论一种在此环境中实现学习的方法,并在信息粒度更高层次的抽象化形式上建立结果模型。知识迁移是一种高效地知识重用,即将源域中收集的现有经验(模型)迁移到目标域中,以支持能量感知的机器学习计算。随后,将进一步讨论知识迁移的被动和主动模式。在这两种模式中,都确定了信息粒度的重要作用。被动方法是在源域的原始模型基础上,在目标域中构建粒度模型,其中原始模型的信息粒度用来量化迁移知识的可信度。主动方法是在目标域中构建一个以损失函数为导向的新模型,其中包含从源域迁移的粒度模型所产生的粒度正则化。构建粒度计算的基本框架有助于方便地解决绿色机器学习的需求,根据模糊集、集合、粗糙集和其他方法对信息颗粒进行概念化的各种方法是有效的解决方案。最后,将概述基于规则架构的面向信息粒度的简洁设计,并在迁移学习和知识蒸馏中采用该设计。

个人简介:

  Witold PedryczIEEE Life Fellow)教授目前担任加拿大埃德蒙顿阿尔伯塔大学电气与计算机工程系教授,《Information Sciences》、《WIREs Data Mining and Knowledge Discovery (Wiley)主编以及Journal of Granular Computing (斯普林格)Journal of Data Information and Management (斯普林格)》的联合主编,是波兰科学院外籍院士和加拿大皇家科学院院士,波兰科学院系统研究所兼职教授。Witold Pedrycz 教授曾获多个奖项,包括 IEEE 系统、人类和控制论协会的 Norbert Wiener 奖、IEEE 加拿大计算机工程奖章、欧洲软计算中心的 Cajastur 软计算奖、Killam 奖、IEEE 计算智能协会模糊先锋奖,以及 IEEE 系统、人和控制论协会2019年优秀服务奖。Witold Pedrycz 教授的研究方向包括计算智能、粒度计算和机器学习等。

                                                                  【编辑:王健】

英文版:

题目:Academic Report Notice of Witold PedryczPursuing Green and Granular Machine Learning: Developments in Federated Learning, Knowledge Transfer, and Knowledge Distillation

Speaker: Academician  Witold Pedrycz

Title: Pursuing Green and Granular Machine Learning: Developments in Federated Learning, Knowledge Transfer, and Knowledge Distillation

Time: 10:00-11:00, September 10, 2022 (Saturday)

Website: https://meeting.tencent.com/dm/ulnDpkMZP8CATencent meeting

(meeting number:482 151 152

Abstract:

  Green Machine Learning (also referred to as Green AI) has recently emerged as an interesting and application-oriented endeavour in the realm of intelligent systems. It stresses a genuine need for a holistic multicriteria assessment of the design practices of Machine Learning architectures and learning schemes by analyzing computing overhead (and associated carbon footprint), interpretability, and robustness, among others. Quite often in real-world environment, data can be available only locally coming with strict constraints imposed on their usage beyond individual data islands. Such restrictions constitute genuine conceptual and algorithmic challenges when it comes to solving problems of global data analysis and the development of global models. A lot of pursuits located in this realm fall under the umbrella of federated learning. We discuss a way of realizing learning in this environment and advocate that the resulting model is built at a higher level of abstraction formalized with the aid of information granule. Knowledge transfer is about a thoughtful and prudently arranged knowledge reuse to support energy-aware Machine Learning computing. Rather than starting from scratch, the existing experience (model) gathered in a source domain is transferred to the target domain. We discuss passive and active modes of knowledge transfer. In both modes, the essential role of information granularity is identified. The passive approach leads to the construction of a granular model in the target domain on a basis of the original model coming from the source domain where information granularity of the model serves as a vehicle to quantify the credibility of the transferred knowledge. In the active approach, a new model is constructed in the target domain whereas the design is guided by the loss function, which involves granular regularization produced by the granular model transferred from the source domain. A generalized scenario of multi-source domains is discussed. Knowledge distillation leading to model compression is studied in the context of transfer learning. We advocate that in order to conveniently address the quest of green machine learning, it becomes beneficial to engage the fundamental framework of Granular Computing. We demonstrate that various ways of conceptualization of information granules in terms of fuzzy sets, sets, rough sets, and others may lead to efficient solutions. To proceed with a detailed discussion, a concise information granules-oriented design of rule-based architectures is outlined. An information granules-oriented design of rule-based architectures in transfer learning and knowledge distillation is used for illustrative purposes.

Personal Introduction:

  Witold Pedrycz (IEEE Life Fellow) is Professor in the Department of Electrical and Computer Engineering, University of Alberta, Edmonton, Canada. He is also with the Systems Research Institute of the Polish Academy of Sciences, Warsaw, Poland. Dr. Pedrycz is a foreign member of the Polish Academy of Sciences and a Fellow of the Royal Society of Canada. He is a recipient of  several awards including Norbert Wiener award from the IEEE Systems, Man, and Cybernetics Society, IEEE Canada Computer Engineering Medal, a Cajastur Prize for Soft Computing from the European Centre for Soft Computing, a Killam Prize, a Fuzzy Pioneer Award from the IEEE Computational Intelligence Society, and 2019 Meritorious Service Award from the IEEE Systems Man and Cybernetics Society. His main research directions involve Computational Intelligence, Granular Computing, and Machine Learning, among others. Professor Pedrycz serves as an Editor-in-Chief of Information Sciences, Editor-in-Chief of WIREs Data Mining and Knowledge Discovery (Wiley), and Co-editor-in-Chief of Int. J. of Granular Computing (Springer) and J. of Data Information and Management (Springer).

                                                             [Editor: Jian Wang]